Civil Aviation Passenger Air Route Preference Prediction Based On Hierarchical Dirichlet Processes

Analyzing the airline choice behavior of civil aviation passengers,identifying the airline preference of passengers and predicting the possible airline choice behavior of passengers in the future.On the one hand,it is helpful for airlines to implement personalized travel product recommendation for different passengers,so as to improve the satisfaction of passengers and the revenue of airlines.On the other hand,it also helps airlines to calculate the route value,plan the route and discover high-value passengers.This paper focuses on the modeling and prediction of airline selection-oriented airline passenger preference behavior,as follows:In view of the group characteristics of civil aviation passenger travel and the hierarchical structure of individual travel intentions,a method for predicting passenger route preferences based on hierarchical travel intentions is proposed.Through statistical analysis of PNR(Passenger Name Record,PNR)data containing passenger historical travel records,extracting the route selection data to obtain the passenger route selection data set;grouping the data set reasonably according to the attribute information containing the geographic origin of the passenger;introducing levels The topic model(Hierarchical Dirichlet Processes,HDP)mines the passenger travel intention and the route selection probability matrix under the travel intention;finally,according to the passenger travel intention,calculates the passenger's future route selection probability,and realizes the prediction of passenger route preferences.The experimental results on the PNR dataset for two consecutive years show that the passenger route preference prediction method based on hierarchical travel intentions has higher prediction accuracy and recall rate than the traditional methods based on popular routes and passenger history selection.On the basis of the passenger route preference prediction model based on hierarchical travel intentions,in order to further solve the problem of sparseness of civil passenger travel data,and based on the idea of collaborative filtering recommendation method,a civil aviation passenger Kangxian preference prediction method incorporating a similarity matrix was proposed.First,we propose a method for calculating similarity between passengers and a method for calculating similarity between routes.The modified cosine similarity is used to calculate the intentional similarity between two passengers,and the passenger similarity matrix is obtained.Use the average of the distance between the two airports to get the similarity matrix between the routes.Based on the original prediction method,the similarity between passengers and similarity between airlines are incorporated.This method is used to predict the passenger's possible route selection preferences in the future.Experimental results on different data sets show that the prediction effect is significantly improved compared with the most commonly used predictions based on popular routes and predictions based on historical route selection records.Compared with the passenger route preference model based on hierarchical travel intentions,the accuracy and recall indicators have been improved to a certain extent.